B9-3, Hi-Tech
Medical Device Park
Wuhan 430206
Tel: +86 27 87880126
Fax: +86 27 87880126



Coronary artery bypass surgery compared with percutaneous coronary interventions for multivessel disease

2016-04-11 / Categories:arterial, hemostasis,
This possibility has been diff  cult to assess because no randomised trial has been large enough to provide adequate statistical power meta-analyses in patient subgroups have been limited by inconsistent reporting in published trials,20 and observational studies have been confounded by treatment selection biases. Pooling of individual patient data from randomised trials substantially increases the number of patients within clinical subgroups of interest and provides a more precise assessment of the eff ects of treatment.21–24 Previous collaborations among clinical trial groups have provided information about variation in the eff  cacy of other cardiovascular treatments according to baseline clinical characteristics.25,26 We undertook a collaborative analysis of data from randomised trials of patients with multivessel coronary artery disease to assess whether the eff ects of CABG and PCI on mortality are modifi ed by patient characteristics.


Patients and procedures

Details of the search strategy that was used to identify relevant trials for inclusion in this collaborative analysis have been reported elsewhere.20 Briefl y, we searched  Medline, Embase, and Cochrane databases for studies published between January, 1966, and August, 2006, by use of terms including “angioplasty”, “coronary”, and “coronary artery bypass surgery”. We also reviewed the reference lists of retrieved articles, conference abstracts, and the bibliographies of expert advisers. We did not limit the searches to the English language. Clinical trials that randomly assigned patients with multivessel coronary artery disease to either CABG or PCI and that reported at least 3 years of follow-up were eligible for inclusion. We excluded trials that compared either method alone with medical therapy, those that compared two forms of PCI, and those that compared two forms of CABG. All included trials were reviewed and approved by ethics committees. We identifi ed 12 eligible trials; the principal investigators of these studies were invited to participate in this collaborative analysis.1–12 Investigators from ten of the trials1–10 provided individual patient data on a set of core clinical variables consisting of demographics (age, sex, and ethnicity), cardiac risk factors (diabetes, smoking, hypertension, and hypercholesterolaemia), clinical manifestations (stable or unstable symptoms, history of myocardial infarction, heart failure, previous PCI, previous CABG, and peripheral vascular disease), angiographic factors (abnormal left ventricular function, number of diseased vessels, and disease of the proximal left anterior descending coronary artery), randomised treatment assignment, and outcomes in follow-up (death, myocardial infarction, stroke, repeat revascular isation, last follow-up contact, and angina). We recoded data from each trial in a uniform format after resolution of data queries and checked data summaries from individual trials against the associated publications for accuracy. The primary outcome measure of this study was all-cause mortality over all available follow-up, and the principal research question was whether survival after random assignment to CABG or PCI was modifi ed by patients’ baseline clinical characteristics.

Statistical analysis

All analyses followed the intention-to-treat principle. For descriptive analyses, we pooled individual patient data from all ten trials and created unadjusted Kaplan-Meier survival curves. For statistical analyses of mortality, we used Cox proportional hazards models stratifi ed by trial24 that included a gamma frailty term to assess random eff ects across the ten contributing trials.27 We tested for interactions of assigned treatment with baseline characteristics by use of multivariable,
stratifi ed Cox models that included treatment assignment, the baseline characteristic of interest, and their interaction.

We also tested the signifi cance of these interactions after including other baseline characteristics in the model.